Precision Agriculture in India: What to Measure First on a 1.08-Hectare Farm

Photo: Shyamli Kashyap / Unsplash

1.08 hectares.

That is the average size of an operational holding in India according to the Agriculture Census for 2015-16, down from 1.15 hectares five years earlier. About 146 million holdings were counted, and roughly 86 percent of them were small or marginal, meaning two hectares or less. Together those farms operate close to half the cultivated land in the country.

Hold that number in your head while you read any pitch for precision agriculture. Most of the pitches were written for farms that are fifty or two hundred times larger.

The rule this post argues for fits in one sentence: on a small Indian farm, measure only what changes a decision this week, and measure water before anything else. Everything below is that rule applied to sensors, radios, satellites, policy and money, with the arithmetic shown so you can check it against your own farm.

Why the Imported Definition of Precision Fails at 1.08 Hectares

The phrase came from large mechanised farms. A combine with a yield monitor and a GPS receiver maps variation across a hundred hectares. A seeder then varies the seed rate by zone. The whole system exists because a tractor driver cannot see variation at that scale, and because the machine is already moving across the field anyway, so an extra sensor on it costs very little per hectare.

None of that transfers. A farmer with one hectare knows every corner of it. She knows where water stands after rain, which edge dries first, where the soil turns heavy. What she cannot see is what is happening twenty centimetres below the surface, what the weather will do in the next three days relative to what her crop needs, and whether the thing she is about to spend money on will pay back this season.

So precision here means something different. It means getting the right amount of an input to the right plot at the right time, where the plot is small, the input is expensive relative to the income, and the decision-maker is one person with a phone. Water, fertiliser, pesticide and labour are the four inputs. Water is the one that sensors measure most directly and most cheaply, which is why this post starts there.

There is a second difference, and it matters more than the first. On a large farm, a sensor system is bought by an agronomy team that will study the data. On a small farm, nobody studies anything. The output has to be an instruction a person can follow without reading a chart: irrigate tonight, hold for two days, spray before Thursday. A node that produces a clean graph and no instruction has not solved the problem it was bought for.

A precision system for a small farm is finished when it produces one sentence a farmer can act on. A dashboard is a by-product, not the product.

The same logic shapes the economics. A farmer deciding whether to pay for a sensor kit is comparing it against a known quantity: the electricity bill for the pump, the diesel for the engine, the cost of one extra spray, the yield loss from one bad week. If the kit cannot be shown to move one of those numbers by more than its running cost in a single season, it will not survive a second season. We have seen this repeatedly in pilots across aquaculture, poultry and dairy work: the technology is rarely what fails. The decision loop around it is.

The Rule: Spend Sensors Only Where a Reading Changes This Week's Decision

Here is the test, stated as four questions. A measurement earns its place only if all four have a specific answer.

1. What decision does this reading change? Not "improves visibility", an actual decision with two or more outcomes.

2. How often is that decision made? Irrigation is decided daily in dry weather. Fertiliser top-dressing is decided two or three times a season. A measurement that is taken hourly for a decision made monthly is wasted battery.

3. What does a wrong decision cost? Over-irrigating a drip line for a day costs electricity. Under-irrigating a flowering crop for three days can cost yield. The cost of being wrong sets how much accuracy you should pay for.

4. Who acts, and how do they receive the instruction? If the answer is "the farmer, by reading a dashboard", the system will fail on the first busy day.

Run the test on the usual candidates and the ranking falls out.

Measurement	Decision it changes	Typical cadence	Who acts on it
Soil moisture at two depths	Irrigate now, wait, or cut the run time	Daily in dry spells	Farmer or pump operator, by SMS or voice
Rain gauge on the plot	Skip tonight's irrigation, delay spraying	Per event	Farmer, automatic with moisture logic
Reference weather (temperature, humidity, wind)	Spray window, disease-pressure alerts	Every few hours	Farmer or advisor
Leaf wetness	Fungicide timing in humid crops	Hourly in the risk season	Advisor, then farmer
Soil electrical conductivity	Leaching irrigation, fertiliser dose	Weekly	Advisor
Full multispectral imaging	Where to walk and look, not what to do	Every few days	Advisor, agronomy team

The first two rows are why this post starts with water. They are cheap, they feed a decision made every day, and a wrong call has a visible cost. The last row is useful, but it is a scouting tool. It tells you where to look. Pretending it tells you what to do is how imaging projects become expensive and quiet.

Notice what is missing from the table: anything that measures a quantity nobody will act on. A node that logs barometric pressure to four decimals is not a precision-agriculture node. It is a data hobby.

Water First: One Acre, One Equation, One Pump

The standard way to estimate how much water a crop uses is the FAO-56 method. Take the reference evapotranspiration for the day, which is how much water a standard grass surface would lose, and multiply by a crop coefficient, Kc, that depends on the crop and its growth stage. The result is the crop's water use in millimetres. Then comes a conversion that makes the whole thing concrete: one millimetre of water over one square metre is one litre.

An acre is about 4,047 square metres. So a crop using 5.75 millimetres in a day uses about 23,270 litres per acre per day. That is the number a farmer can compare against her pump.

Efficiency then enters. Drip systems deliver most of what they pump to the root zone, and flood irrigation loses a large share to percolation and runoff. The code below turns the arithmetic into a function and runs it on a worked example. The inputs are illustrative: a reference evapotranspiration of 5 millimetres a day, which is plausible for a warm dry-season day in much of peninsular India, and a mid-season crop coefficient of 1.15, which is within the range used for many vegetable and field crops. Your own values should come from your crop, season and nearest weather station.

ACRE_M2 = 4046.8564224

def daily_need_litres(et0_mm, kc, rain_mm, area_m2, efficiency):
    # FAO-56 style: crop water use = Kc x ET0; 1 mm over 1 m2 = 1 litre
    etc_mm = kc * et0_mm
    net_mm = max(0.0, etc_mm - 0.8 * rain_mm)   # count 80 percent of rain as effective
    return net_mm * area_m2 / efficiency

def pump_hours(litres, pump_lph):
    return litres / pump_lph

print(round(daily_need_litres(5.0, 1.15, 0.0, ACRE_M2, 0.90)))   # 25855 litres, drip
print(round(daily_need_litres(5.0, 1.15, 0.0, ACRE_M2, 0.60)))   # 38782 litres, flood
print(round(daily_need_litres(5.0, 1.15, 10.0, ACRE_M2, 0.90)))  # 0 after a 10 mm rain

The first two results are the ones to dwell on. Run on one acre, the same crop on the same day needs about 25,900 litres delivered by drip and about 38,800 litres delivered by flooding. That is a 33 percent gap in pumped water for identical crop use. The third line shows the effect a rain gauge has on the decision: after 10 millimetres of rain, counting 80 percent as effective, the crop's whole day of water use is covered and the right amount to pump is zero.

Method	Field efficiency used	Litres per acre per day	Pump hours at 6,000 L/h
Drip	90 percent	about 25,900	about 4.3
Sprinkler	75 percent	about 31,000	about 5.2
Flood	60 percent	about 38,800	about 6.5

Those efficiencies are round planning figures, not measurements of any particular system. The pump rating of 6,000 litres per hour is an assumption for the example. The point is the structure: the same crop demand turns into different pump hours depending on how the water is delivered, and a sensor that tells you when the root zone is already wet lets you stop pumping earlier than a timer would.

Pump hours are the unit a farmer understands. Translate every water number into hours of pump time and electricity, and the conversation about sensors becomes a conversation about the bill.

Why do you need soil moisture at all if the equation gives the answer? Because the equation gives the average crop demand and the farm is not average. The sandy corner drains in a day, the clay corner holds water for a week, and rain in a monsoon gap is patchy at the scale of a village. The equation sets the budget. The soil sensors tell you whether the budget is landing where the roots are.

What a Soil-Moisture Node Has to Survive

The sensors in this class are capacitive probes. They measure the dielectric constant of the soil around the probe, which changes strongly with water content, and convert it to a volumetric water content, written VWC and quoted as a percentage. They are cheap, they draw little power, and they do not corrode the way old resistive probes did. They also come with three habits that bite in the field.

First, the factory calibration is for a generic soil. Black cotton soil, red soil and sandy loam read differently for the same wetness. A usable deployment records a field calibration for each soil type: take a dry reading, saturate the soil around the probe, let it drain to what the farmer calls "right", and note the three values. Without that, the numbers are good for trends and poor for thresholds.

Second, salinity and temperature leak into the reading. In irrigated soils with salt build-up, a capacitive probe reads wetter than the soil is. Temperature shifts the reading by a small amount for each degree. The fix is to log soil temperature beside moisture and to treat any step change after fertiliser application with suspicion.

Third, installation decides more than the sensor does. An air gap along the probe body, which opens when the soil dries and cracks, makes the reading drop sharply. A probe pushed into a stony layer breaks. A probe near the dripper reads wet all season. Burying the probe in a slurry of the same soil and tamping it, then placing it between two emitters at the edge of the wetted bulb, produces readings that track the crop.

Two depths are the minimum useful layout. A probe at about 15 centimetres tracks the active root zone for shallow-rooted vegetables. A second at about 30 centimetres shows whether irrigation is reaching deep enough or whether water is draining past the roots. The gap between the two tells you more than either alone: a wet shallow reading over a dry deep reading means you are watering often and lightly, and the crop's roots will stay shallow.

Design item	What to choose	Why
Probe depths	15 cm and 30 cm	Root-zone moisture plus a drainage check
Probe count	Two nodes per acre in uniform fields, one per soil zone otherwise	Soil variation, not field size, sets the count
Housing	Sealed enclosure rated for dust and temporary immersion, cable gland on the underside	Monsoon water and irrigation spray reach every joint
Power	Primary lithium cell sized for a multi-year life at the chosen reporting interval	No mains, no one to recharge, and a dead node reads as a dry field
Reporting interval	10 to 30 minutes in dry spells, longer in cool weather	Matches the daily decision, not the minute
Local rule	Node flags itself when the reading leaves a plausible range	A broken probe must not trigger a pump

The last row deserves a sentence of its own. A node that reports a reading of zero because the cable has been chewed by a rodent looks identical, at the server, to a field that has gone completely dry. If a pump is wired to that reading, it runs all night. Every node should send a status byte beside its readings, and every automation rule should require both the reading and a healthy status before it acts.

Getting Eight Bytes Out of a Field with No Mains and No Network Guarantee

The data from a node is tiny, and the way you pack it determines the battery life. Here is a layout that fits a battery reading, two moisture values and a soil temperature into eight bytes, with a round-trip test.

import struct

# big-endian: battery mV (u16), VWC at 15 cm and 30 cm in 0.1 percent (u16 each), temperature in 0.1 C (i16)
def pack(batt_mv, vwc15_pct, vwc30_pct, temp_c):
    return struct.pack(">HHHh", batt_mv, round(vwc15_pct * 10), round(vwc30_pct * 10), round(temp_c * 10))

def unpack(b):
    mv, a, c, t = struct.unpack(">HHHh", b)
    return mv, a / 10, c / 10, t / 10

p = pack(3612, 27.4, 31.8, 29.6)
print(p.hex(), len(p), unpack(p))   # 0e1c0112013e0128 8 (3612, 27.4, 31.8, 29.6)

Eight bytes of application data plus the 13 bytes of LoRaWAN framing make a 21-byte frame. At spreading factor 9 and 125 kHz bandwidth, the radio is on air for about 185 milliseconds per uplink, by the standard airtime formula. A node that reports every ten minutes sends 144 frames a day, which comes to about 27 seconds of airtime across the day, a duty cycle of roughly 0.03 percent. The licence-exempt rules in India for LoRa-class devices in the 865 to 868 MHz band, set by the Department of Telecommunications in G.S.R. 853(E) of 10 December 2021, cap duty cycle and power by device class, and this load is far inside any of them. The wireless equipment itself needs the type approval the rules require.

That arithmetic is the reason LoRa suits this job. A multi-kilometre line of sight across open farmland lets one gateway on a pole or rooftop serve a cluster of villages' worth of plots. Our LoRa range work on Indian farms, covered in a separate field guide, shows how quickly range collapses once a node sits low in a crop canopy behind a bund or a row of coconut palms, so the gateway placement and the node height deserve a half-day on site before any purchase.

The failure that matters at this stage is not radio. It is silence. A node whose battery dies, whose antenna is crushed by a tractor wheel, or whose gateway loses power looks the same as a node with nothing to say. Treat a missing report as a first-class event: if two consecutive uplinks fail, the server marks the plot "unknown" and the advice becomes "water by the calendar, and send someone to check the node", never "assume no change".

A sensor that goes quiet must read as a warning, never as a reassurance. Design the silence handling before you design the dashboard.

Satellites, Drones and Ground Nodes: Which One Answers Which Question

Ground nodes measure one point continuously. Satellites measure every point occasionally. Drones measure a chosen area on demand. Each answers a different question, and mixing them up is the commonest way to waste money in this field.

Source	Typical revisit	Ground detail	Cloud sensitivity	Best question it answers
Optical satellite (10 m class, public data)	About every five days or better, cloud permitting	10 m pixels, so a one-hectare plot is about 100 pixels	High, which matters in the monsoon	Which plots look stressed relative to their neighbours?
Radar satellite (C-band, public data)	Roughly 6 to 12 days depending on region and constellation status	Tens of metres	Low, it sees through cloud	Is the field flooded, ploughed, or growing, even in the monsoon?
Drone with multispectral camera	On demand, limited by weather and a pilot	Centimetres	Low, needs daylight and light wind	Why does this part of this plot look different?
Ground moisture node	Every 10 to 30 minutes	One point at two depths	None	Do I irrigate tonight?

The pattern that works is to let the satellite choose where to look and the ground node decide what to do. An optical index such as NDVI, computed from public Sentinel-2 imagery, will not tell a farmer to irrigate. It will show that one plot has dropped relative to its neighbours over ten days. That is the cue to walk there, or to send a drone, or to check that plot's moisture node. The satellite is a cheap, wide, regular scout. The node is the decision instrument.

Two cautions apply on a one-hectare farm. First, a ten-metre pixel mixes the crop with bunds, paths and neighbours' fields at the plot edges, so indices near boundaries lie. Using only interior pixels, or comparing a plot against its own history rather than against other plots, removes most of the error. Second, in the kharif season the clouds that bring the rain also blind optical satellites for weeks. Radar fills the gap for questions about water and field state, though it cannot give a simple greenness number.

EurthWatch, the satellite-intelligence line we are building, is built on exactly this division of labour: public optical and radar data for the wide view, with ground sensing used to ground-truth it. The ground-truth part is the one most pilots skip and the one that decides whether anyone trusts the map.

What the Digital Agriculture Mission Gives You to Build On

Policy matters here for a practical reason. A sensor platform that can attach its readings to an official plot identity is more useful than one that cannot, because advisories, credit and insurance all key off the same identity.

The Union Cabinet approved the Digital Agriculture Mission on 2 September 2024 with an outlay of 2,817 crore rupees, of which the Centre's share is 1,940 crore. It has two pillars. AgriStack is the registry layer, made up of a farmers' registry, geo-referenced village maps and a crop-sown registry. The Krishi Decision Support System is the analytics layer, built on geospatial data and used for things like drought and flood monitoring. A Farmer ID, called the Kisan Pehchaan Patra, is the Aadhaar-linked key that ties a person to land records.

As publicly reported through October 2026, the Ministry announced that the target of 11 crore Farmer IDs had been met on 23 October 2026, five months ahead of the 31 March 2027 deadline. The Digital Crop Survey, which records what is actually grown on each plot, was reported to have covered 648 districts and more than 31.3 crore plots in the 2025-26 rabi season. One reported caveat is worth keeping in view: in August 2026 the Agriculture Ministry said roughly 40 percent of the sown area was yet to be linked to Farmer IDs. These figures change monthly, so check the Ministry's own pages before you quote any of them.

What does this do for a small-farm sensing project? Three things.

It supplies a plot boundary that is not drawn by the vendor, which lets satellite analysis be tied to a defined polygon. It gives a route to link a farm's records to credit and scheme access, which is often the strongest argument for a farmer to share data at all. And it points to a likely future in which advisories are delivered per plot, in the local language, through channels the farmer already uses.

It also raises a question the sensor business has to answer plainly: who owns the readings? The Digital Personal Data Protection Act, 2023 and its Rules, notified on 13 November 2025, are being phased in, with the core operational duties scheduled to start around May 2027. Whether a particular moisture reading counts as personal data depends on how it is tied to a person or a holding, which is a legal question for your counsel, and nothing here is legal advice. The engineering answer is simpler: keep the reading, the plot identity and the person's contact details in separate stores with separate access rules, so that any later interpretation can be satisfied without rebuilding the system.

Three Acres, Two Seasons: A Representative Composite

The following is a representative composite, genericised, assembled from several pilots in drip-irrigated vegetable plots. The numbers are illustrative of what such a deployment looks like, not a measured result from any named farm.

The plot is three acres of mixed vegetables on two soil types, black clay in the lower half and a lighter red loam on the rise. The farmer ran drip irrigation on a timer: 90 minutes morning and evening, adjusted by feel. Electricity for the pump was the largest recurring cost after labour.

The first design mistake was obvious in hindsight. Six nodes went in, two per acre, spaced evenly. Four ended up in the clay and two on the rise, and the two on the rise read drier than the farmer's own judgement for the whole first month. It took a walk with the farmer to find out why. The rise was where she irrigated longest by hand-feel, and the probes were sitting in a spot that drained within hours of a run. The nodes were right. The timer was wrong for that zone.

The second mistake was the one that costs adoption. The system sent a daily dashboard link. Nobody opened it. After three weeks the logic changed to a single message in Telugu, sent at 5 pm: "Rise plot: dry at 30 cm, water tonight 60 minutes. Clay plot: wet, skip." The farmer acted on it the same evening, and kept acting on it.

Over the next two seasons, with the evening message driving a split schedule by zone, the pump run time fell by an illustrative 20 to 25 percent in dry weeks, with no visible yield penalty on the clay plot and a modest yield gain on the rise, where the old timer had under-watered in the heat of April. The saving that mattered most to the farmer was not the water. It was the three evenings in a month when she no longer drove out to check.

The failure mode in the second season is the instructive one. Salt had built up near the drippers on the clay, and the 15 centimetre probe began reading wetter than the soil was. Without a temperature and conductivity check, the node would have kept saying "skip" on a plot that needed a leaching irrigation. A field check with a handheld meter caught it, and the node's rule gained a second condition: if VWC stays high while the 30 centimetre reading falls, flag the plot for a manual check.

The first season of a sensor deployment is spent finding out where your assumptions about the farm were wrong. Budget for that in the plan, because it is the main product of season one.

Where This Is the Wrong Answer

Sensors are the wrong first purchase in several situations, and saying so plainly is part of doing this work honestly.

If the farm has no controllable irrigation, there is nothing for the reading to change. A rain-fed plot with a hand-held bucket needs better weather advice and variety choice, not a probe.

If the pump is shared and scheduled by a cooperative or a village committee, the farmer cannot act on a plot-level message. The decision unit is the committee, and the useful product is a village-level water schedule.

If the farmer's phone cannot receive the message in a language and format she uses, the system fails at the last metre however good the nodes are. Test the message channel before you test the sensor.

If the crop is low in value per acre and the water is cheap, a sensor may never repay itself. That is arithmetic, not pessimism. Run the pump-hours calculation above, price the electricity or diesel in your own district, and compare the saving to the annual cost of the kit and its support. If the saving is smaller, do not deploy.

And if nobody on the supplier side will be present in month four, when the first node fails and the farmer calls, do not start. Field electronics need maintenance, and the cost of a support visit has to be in the price from the first day.

Starting Next Week: Four Steps, One Decision at a Time

1. Pick the decision. Choose one: irrigate or hold. Write down who makes it, when, and what they use today. If you cannot name the person, stop.

2. Walk the plot with the person. Mark the soil zones, the wettest and driest corners, the pump and its controls. Place nodes by soil zone, not by grid.

3. Calibrate and baseline. Take dry, saturated and field-capacity readings per soil, and run for two weeks with no change to the farmer's routine. The baseline is what makes any later saving believable.

4. Turn on the one-sentence message. Send a daily instruction in the farmer's language and channel. Review after four weeks: did she act, did it help, which readings were wrong?

Only after a season of that is it sensible to add the weather station, the satellite scouting and the leaf-wetness probe. Each addition has to pass the four-question test again.

For anyone buying or commissioning such a system from a vendor, a short list of questions saves money. What is the field calibration procedure, and who does it? What happens when a node goes silent? What is the status byte, and does any automation require it? Who is on site in month four? Which standard and approval cover the radio? Can I export my data, and in what format? A vendor who answers these in specifics has done this before.

Back to 1.08 Hectares

The farm in the opening paragraph is not an edge case. It is the median. A precision-agriculture system for India has to work at that size, for that farmer, with that phone, and pay for itself inside a season.

The sequence is short enough to remember. Water before everything else. Two nodes per soil zone, calibrated by hand. One message a day in the farmer's own language. A satellite as scout and a node as decision-maker. Silence treated as an alarm. Only the measurements that change this week's decision.

Everything else is a chart.

Related service: AgriTech and Precision Agriculture

Related case study: Intelligent Aquaculture Hub with Edge AI

EurthTech is an engineering-first product design and embedded systems company with a development center in Hyderabad. To discuss a project, write to connect@eurthtech.com or see https://www.eurthtech.com/case-studies.